Vous êtes sur la page 1sur 12

Improvements in intra-site spatial analysis

Franois Djindjian*
Resmee franaise
Aprs un rapide historique des mthodes d'Analyse Spatiale de l'Habitat, replaces
dans le contexte de l'volution des techniques statistiques, une nouvelle methode
d'analyse spatiale est propose, comme une amlioration de la methode de R. Whallon
'Unconstrained clustering'.
La nouvelle mthode, intitule 'Analyse de la structure spatiale de l'Habitat', com-
prend les tapes suivantes:
enregistrement des distributions spatiales, par coordonnes de points, ou comp-
lissage des distributions et traitement des donnes manquantes,
chantillonnage d'un tableau de vecteurs de densit,
analyse des correspondances du tableau de vecteurs de densit et slection des
facteurs rvlant des structures spatiales,classification ascendante hirarchique,
soit classique, soit sous contrainte de contigut, sur les facteurs,
visualisation et caractrisation des structures spatiales.
La mthode a t applique au site esquimau 'Mask' (L. Binford), l'habitat mag-
dalnien de Pincevent (A. Leroi-Gourhan) et au village nolithique lacustre de Charavtnes
(A. Bocquet). , , , ^ , , . ,
Les rsultats des exprimentations sont enfin compares, et la valeur de la methode
6.1 Introduction
This paper describes the preliminary results of methodological work concerning the
definition of an Intrasite Spatial Analysis method, the choice of data analysis algorithms
CNRSMusee des Antiquits Nationales,
32 Rue Boussingault
75013 Paris
and associated software, and the experiment on archaeological data to determine
possibilities and limits of the method in comparison with the archaeological aim and
the other intrasite spatial analysis methods.
6.2 The main existing nnethods of intrasite spatial analysis (ISA)
The study of the historical development of Intrasite Spatial Analysis methods allows
the identification of three technical phases:
1. The first period (1950-1975) saw the influence in archaeology of quantitative
ecology techniques, which appeared during the fifties, and which was applied
to archaeology during the seventies (Whallon, Hodder); the applications were
based on spatial distribution tests either on counting data (dimensional analysis
of variance), or on coordinate data (nearest neighbour analysis).
2. The second period (1970-1980), an extension of the previous period, is based on
association tests between spatial distributions. The best element of this set of
techniques is the multi-response permutation procedures (Berry et al. 1984).
3. The third period saw the coming out after 1975 of data analysis techniques in
intrasite spatial analysis:
local density analysis (Johnson 1977, Johnson 1984),
cluster analysis on spatial coordinates (Kintigh & Ammermann 1982; Simeck
& Larick 1981),
correspondence analysis (Hesse 1984, Bouchet 1986), unconstrained cluster-
ing (Whallon 1984).
Hodder and Orton's book (Hodder & Orton 1976) and Orton's paper (Orton 1982a)
give a good introduction, mainly for the the techniques of the first two periods. The
book edited by Hietala (Hietala 1984), gives a good but non complete methodological
review, and a set of archaeological applications.
6.3 Comparison of intrasite spatial analysis methods based on data
analysis techniques
The local density analysis by Johnson (Johnson 1977, Johnson 1984) is based on the
definition of an association index between two spatial distributions. The index matrix is
then processed by data analysis techniques (multi-dimensional scaling in the first paper,
correspondence analysis in the 1985 paper), which give a reduction of the distribution
space (R-method).
Graham (Graham 1980) emphasized the impropriety of the association index, which
is strongly linked with the pattern of the distribution: sometimes the value of the
association index between two different distributions is greater than the value of the
association index between the same distributions. The association index, based on
the nearest neighbour, as proposed by Graham (Graham 1980), avoids the previous
disadvantage, but the value of the association index is always dependent of the pattern
of each distribution. The two methods, moreover, are R-methods, which give only a
reduction of the space of distribution, and not a pattern of the archaeological area.
Cluster analysis on spatial coordinates (Kintigh & Ammermann 1982), is not a multi-
dimensional spatial analysis technique, even if the method uses cluster analysis. The
multi-dimensional approach is only given by a graphical superposition of the cluster
variance circles. The method, more oriented to the research of superposition of artefact
concentrations, is not, sensu stricto, an intrasite spatial analysis method.
Unconstrained clustering (Whallon 1984), is a method based on the computing, in a
point of the area, of a multidimensional density vector. Results of the cluster analysis
applied to the density matrix give a spatial pattern. Interpretation of clusters may
induce the definition of specialized activities areas. The method, unfortunately, is a
Q-method, which gives only a pattern of the archaeological area.
6.4 A new method: analysis of the intrasite spatial structure (ISS)
6.4.1 Critical examination
The critical examination of the existing spatial methods induces the following conclud-
ing remarks:
Data analysis techniques supply a better solution to the problem of intrasite spatial
analysis, in comparison with distribution tests, or association tests between two
Data analysis techniques, nevertheless, need the following features:
- the capacity for producing a R+Q multidimensional reduction (both on vari-
ables and points or cells),
- the capability to be a non-linear multidimensional technique, as already
pointed out by Whallon (Whallon 1984), concerning the problematic use of
principal components analysis.
Smoothing techniques allow simultaneous use of spatial data obtained both by
cell counting and by coordinates measures.
Informations of spatial distribution patterns (clustered, random or uniform dis-
tribution) must be keep apart from informations given by association between
spatial distributions, by the following manner:
- Multidimensional approach of intrasite spatial analysis is best explained by
the definition, in a point of the area, of a multidimensional density factor.
- Best structuring of the spatial area is obtained in considering also the infor-
mation given by the contiguity of density vectors.
A new method called the the 'analysis of the intrasite spatial structure', which integrates
the previous technical features, may be seen also as a structuration method in Archaeol-
ogy (Djindjian 1980), especially in the main point of the choice of intrinsic informations
(here the distributions), directly connected with the spatial analysis problematic.
The ISS method is constituted by the following steps:
1. Recording of spatial distributions, either by point coordinates or by cell counting,
2. Smoothing of spatial distributions (moving average, Symap... ),
3. missing data processing (smoothing, krigeing,... ),
4. Sampling of a density vector matrix,
5. Correspondence analysis on the density vector matrix,
6. Selection of factors supplying relevant spatial structures,
7. Hierarchical ascendant clustering applied on previous factorial coordinates,
8. Option: hierarchical ascendant clustering with contiguity constraints. In that
special case, the method is called the analysis of the intrasite contiguous spatial
structure (ICSS).
9. Visualization and characterization of the structures.
6.5 Technical improvements in tlie ISS method
Without considering the missing data processing which need not here particular ex-
planations, improvements of Whallon's unconstrained clustering are defined by the
introduction of two data analysis techniques.
6.5.1 Correspondence analysis
Discovered by Jp. Benzecri (Benzecri 1973) about twenty years ago, correspondence
analysis has been introduced by Djindjian (Djindjian 1976) in archaeology, involving
many applications in French archaeology (described briefly in section 6.7).
Correspondence analysis is employed here as a factor analysis for the multidimen-
sional reduction of a density vector matrix. R+Q visualization allows to represent on
the same map both the distributions and the sampling points. Non-linear analysis by
the chi square distance, correspondence analysis can recognized local patterns, on the
contrary of a linear analysis, Hke PCA, which shows trend patterns.
6.5.2 Hierarchical ascendant clustering with contiguity constraints
The cluster analysis, used here, is a hierarchical ascendant clustering algorithm based on
chi square distance and intercluster variance maximizing aggregation.^ The algorithm
is applied not on the original density vector matrix, but on the coordinates of the
sampled points in the selectioned factorial axis.
The hierarchical ascendant clustering with contiguity cor\straints is a specialized
algorithm for applications in human geography. The algorithm allows to two points to
be clustered only if they are spatially contiguous. Clusters, then, are both multidimen-
sional density clusters and spatial clusters. An algorithm has been realized by Lebart
(Lebart 1978), based on the recording of a contiguity table updated at each step of the
algorithm, like the distance table. The same approach has been employed successfully
^Variance criteria are unanimously considered by the French school of data analysis as a better criterion
than the archaic criteria of cluster analysis (single linkage, ward... ).
in archaeology with a toposeriation method (Djindjian 1985) using a non-hierarchical
clustering algorithm, to show the spreading steps of a cemetery, from both a senation
and a grave contiguity table.
6.6 Case studies
Three case studies has been chosen for the experimentation of the new method The
two first data sets, the Nuniamut Mask site from L. Binford, and the Magdaleman
Pincevent camp-site excavated by A. Leroi-Gourhan, are now reference data sets for
intrasite spatial analysis method tests. The last example, the Neolithic lacustrine vlage
of Charavines from A. Bocquet is tested to verify the capability of the method on a
Neolithic site.
6.6.1 Intrasite spatial analysis of the Nunamiut Mask site
The Mask site has been studied by L. Binford (Binford 1978), for the comparison
between the archaeological interpretations from the artefact spatial distributions, and
the real Nuniamut activities. The data published by Binford have been processed by
Whallon (Whallon 1984) as a case study for the 'unconstrained clustering' method. 1 he
five spatial distributions are tools, projectile components, wood scrap, bone scrap, large
bones They show strong clustering. Correspondence analysis shows on the four axis
an individual behaviour of each distribution:
axis 1: Tools
axis 2: Association of projectile components and wood scrap in opposition with
bone scrap
axis 3: Opposition between projectile components and wood scrap
axis 4: Large bones
The activity areas are defined by only one artefact type, showing that the case study
is a caricatural example, not able to be used to shidy the limits of different methods.
Three methods (UC, ISS, ICSS) supply, therefore, the same number of clusters (six)
and the quite same spatial patterns (Fig. 6.1). The only practical conclusion of the
experiments is the capacity of the three methods to recognize very clean activities area.
6.6.2 The Magdalenian camp site of Pincevent (area 36)
The publication of the 'area 36' of the famous Magdalenian camp site of Pincevent (A
Leroi-Gourhan and M. Brezillon (Leroi-Gourhan & Brezillon 1972) supphes a classical
but very detailed spatial analysis of the occupation floor. Data sets have been used
as a case study for the experimentation of their methods by Johnson Oohr^on 1977,
Johnson 1984) for local density analysis and by Kintigh & Ammermann 1982 for the
analysis on spatial coordinates. , , n
The twelve spatial distributions are lithic tools (endscrapers, burins and burin spalls,
awls backed bladelets), bone tools, cores, hearth stones and reindeer skeletal remains
(five'types). Most of the spatial distributions have a strong clustering. Correspondence
6 fi fi
S S 5 S t fi < .

S 5 5 s s 6 ( fi
S S S S s s fit
4 fi fi fi
3 3
S 3
S 5 S S
fi fi s 5 s t t fi
fi fi ( . . t fi . . .
2b - I S s
I Tooll : T
1 Prajvctil : Tirgct ihootlnq icttvltlti
4 Hood icrip : Cnft lr4i
5 Tooll, lirgc bones ind bone script :
fin! proccsiInq ind
COnsuaptlon (cttvltlci
Z Bant icrtp : Discard ir

fi t fi
t 1 t
fi ( s
fi fi
4 s s
t ? f I S S
fi 1 i s s s
J 3 4 s s
1 3 S S
3 3 3 S S

I 5 S
i Z % i
I I s i 5
t z z 5 s s
fi 3 Z S 5 5
3 3 * 5 S
3 3 3 S 5
3 3 I S S
6 fi
1 fi (
t (
6 i
S t
fi t fi .
Figure 6.1: The Nunamiut 'mask' site
analysis shows an opposition between bone and lithic tools, core, hearth stones and
faunal remain (Fig. 6.2a). The lithic tools shows a same behaviour in the reduced
factorial space, i.e. similar spatial distributions, structure confirmed by the results of the
cluster analysis on spatial coordinates which show superpositions of cluster variance
circles (Fig 6 2b) The faunal remains set, also clustered, shows nevertheless some
variability among the first axis, characterizing a differential dispersal from the central
area The results are globally similar with the results supplied by the local density
analysis (Fig. 6.2c). Cluster analysis shows a spatial structure, characterized by a tool
cluster in the central area, around the hearths, a strong differential disper^l of faunal
remains, and a globally uniform distribution of small hearth stones (Fig. 6.2d, e).
The interpretation of the spatial patterning cannot induce to specialized activity areas,
but rather to a central area of non specialized activities, perhaps only resulting from
a smoothing due to a long occupation time. Site maintenance activities may explam
the dispersal area structures. Intrasite spatial analysis, realised on the whole 36 area
of Pincevent, gives finally unsatisfying results. A complementary analysis, at a more
detailed scale, is necessary to try to discover patterns inside the central area.
The main conclusions of the experiment are the strong sensitiveness of the methods
to artefact dispersal patterns, directly associated with density measures, and the relative
insensitiveness to smoothed patterns, which seem to limit the application of mtrasite
spatial analysis method to short occupation sites.
6.6.3 The Neolithic lacustrine village of Charavines
The Neolithic lacustrine village of Charavines (Isre) has been occupied twice during
the late neolithic. The site, excavated by A. Bocquet since 1972 is not yet published.
Processed archaeological data correspond to a 10 x 15 m^ Umited area from the firs
occupation floor (B3). The twelve spatial distributions are: large flint pieces, smal
flint pieces, fine ceramics, rough ceramics, quartzites, stones, charcoal, clay, nuts, burnt
bones, coproliths, cereals. Data are counted or weighed by metric triangles.
Correspondence analysis, realised here on complete disjunctive coded variables,
shows cor^piex associations between the modaUties of variables and Guttman effect.
theTwo first axis, with flint, ceramics, bones and nuts associations are mterpreted
as human activities. The third axis, with stones, clay and charcoal associations, is
interpreted as building material remains.
Cluster analysis shows a three cluster structure and a five cluster structure, visualized
by^6 3. The comparison with the restored map of the houses from wood posts
dated by dendrochronology, clarifies the interpretation of the clusters Cluster D
CO responds to the outside of the houses, and clusters A, B, E to the inside, while
c'l^'erC corresponds probably to a porch or an enclosure alongside the house. Inside,
clSers A and B are acdvity areas, around the hearth, while cluster E is more probably
a rest area or a circulation area. < ^u
The results of the spatial analysis find, spectacularly, the house map confirming the
map obtained by dendrochronology. The determination of activities areas inside ^
ne?ertheLs too vague, to be related, perhaps, with the small number of variables, or
their limited relevance.
^The spatial analysis has been re^Ii^^Iunder our direction by A.M. Christien in a work carried out at
the University of Paris X Nanterre.
s - Corres pondence analys is of local dens ity
as s ociation indices ( J O H N S O N , 1 9 8 4 , fig . 5 )
1. Antlers and fragments
2. Haxill iaries
3. Mandibles
4. Isolated teeth
5. Ribs
6. Scapulae and pelvises
7. Humeri
8. Femurs and rotules
9. Tibias
10. U1nas and radii
19/20 25)
21 y
Axis 1
Axis 2
7 /10 \
b - Corres pondence analys is of dens ity vector matrix
11. Metatarsals 16. Hearth stones (+ 7 cm)
12. Hetacarpals 17. Hearth stones (- 7 cm)
13. Metapodials 18. Cores
14. Phalanges 19. Burins
15. Hare bones 20. Burin spalls
21. Backed bladelets
25. WorWed bones 22. Awl s
23. Endscrapers and truncated blades
24. Artefacts in brown flint
too I 102 103 KM IS KM 7 KM 109 110 III III 111 IM IIS IM 117 IIS llf
General plan of Area 36
(from JOHNSON, 1985, fig. 3)
d - Spatial coordinate cluster Analysis:
Clusters for mandibles, metacarpals and
metatarsals (from KINTIGH-AMMERMAN, 1982, fig. 8a)
3 4
2 3
4 4
3 3
3 2
2 3 4
4 3
3 2

3 2
2 2 2
e - ISS on area 36
Cluster 4: 40 hearth stone, 30% tools, 30% faunal remains
Cluster 3: 351 hearth stone, 5% tools, 60% faunal remains
Cluster 2: 50% hearth stone, 50% faunal remains,
essentially ribs
Figure 6.2: The Magdalenian camp site of Pincevent
Ea B
4a - I S S (3 clusters)
4b - Wood post map
Cluster A: burnt bones - fine ceramic
Cluster B: nuts, cereals, ceramic, stones, charcoal
Cluster E: less artefacts than B, A
Cluster D: without structure
Cluster C: quartzite, coproliths, no clay, no charcoal
4c - I S S {5 clusters)
Figure 6.3: The Neolithic lacustrine village of Charavines
6.7 Conclusions
The experimentations on the three case studies demonstrate the following conclusions:
The ISS method realizes an intrasite spatial analysis more efficiently than the
local density analysis, or the cluster analysis on spatial coordinates, and more
analytically than the unconstrained clustering.
The ISS method proves its capacity to recognize strong spatial structures (ex-
amples 1 and 3), but its limits are unknown when applied to fuzzy structures
(example 2).
Correspondence analysis confirms to be an essential tool, before the use of a
cluster analysis, in order to understand the spatial variability of the data.
Nevertheless, it seems untimely to confirm the superiority of the cluster analysis
with contiguity constraints, on the basis of the results of the first example.
If we can consider these results are encouraging, new analysis on the same data
and on other data are necessary before concluding these methods are definitively
operational for intrasite spatial analysis.
Appendix: some references of applications of correspondence analysis
in French Archaeology
Djindjian 1977 (paleolithic burins); Dive 1984/Dive 1985 (acheulean retoucUed tools);
Boutin, Tallur, Chollet 1977 (azilian points); Vigneron 1979 and Decormeille, Hinout
1982 (mesohthic points); Buret 1985 (polished axes); Julien 1982 (harpoons); Wuaillat
and Massonie 1974 (Bronze axes); Moscati 1986 (Etruscan mirrors); Leredde and Perin
1980 (merovingian 'plaques-boucles')
Djindjian et DeCroisset 1976 (mousterian handaxes); Tuffreau et Bouchet, 1985 (acheulean
handaxes); Giligny 1986 (neolithic ceramics); Mohen 1980 (iron age ceramics); Courbin
1983 (greek ceramics)
Structuration of industrial facies
Djindjian 1977 a 1986 (aurignacian); Djindjian 1981 (gravettian); Djindjian 1988 (mag-
dalenian); Hours 1977 (Middle-East epipaleoUthic); Decormeille et Hinout 1982 (mesolithic);
Slachmuylder 1984 (mesolithic), Mohen et Bergougnan 1984 (neolithic)
Audouze 1984 (bronze age pins); Djindjian 1980 (Munsingen!); Djindjian, 1985 (merovin-
gian cemetery); Desachy et Bourgeois 1987 (greek cemetery); Brunet et alii, 1985 (bronze
and iron age cemetery)
Intrasite spatial analysis
Bouchet 1985; Hesse 1984
6.7.1 Paleoecoloy
Djindjian et Marquet 1985 (rodents); Laurin, Rousseau 1985 (molluscs); Villette 1984
(birds); Blanc-Vemet 1976 (foraminifera); Ozouf 1983 (sedimentology); Gasse 1980
(diatoms), etc...
BENZECRI, J. P. 1973. L'Analyse des donnes. I. La Taxinomie, H. L'Analyse des correspon-
dances. Dunod, Paris.
BERRY, K. J., K. L. KVAMME, & P. W. MIELKE 1980. "A permutation technique for the
spatial analysis of the distribution of artefacts into classes", American Antiquity, 45:
BERRY, K. J., P. W. MIELKE, & K. L. KVAMME 1984. "Efficient Permutation Procedures
for Analysis of Artifact Distributions", in Hietala, H. J., (ed.), Intra-Site Spatial Analysis
in Archaeology, pp. 54-74. Cambridge University Press, Cambridge.
BiNFORD, L. R. 1978. "Dimensional analysis of behavior and site structure learning
from an Esquimo hunting stand", American Antiquity, 43: 330-361.
BOUCHET, J. P. 1986. Les gisements du palolithique moyen de Biache st Vaast et Seclin :
organisation spatiale des vestiges archologiques. Diplme de l'EPHE, Ille section.
DJINDJIAN, F. 1976. Contributions de l'analyse des donnes a l'tude de l'outillage de pierre
taille. Universit de Paris 1. 2 vol.
DJINDJIAN, F. 1980. Construction de systmes d'aide la connaissance en archologie
prhistorique: structuration et affectation. PhD thesis. Universit de Paris I. 2 vol.
DJINDJIAN, F. 1985. Seriation and toposeriation by correspondence analysis, pp. 118-135.
GRAHAM, 1. D. 1980. "Spatial analysis and distance methods in the study of archae-
ological distributions". Journal of Archaeological Science, 7: 105-129.
HESSE, A. 1984. La rpartition cramique la surface de l'enclos SAV2 de l'Ile de Sa d'aprs
l'analyse des correspondances, 7. Cripel, Lille.
HIETALA, H. J., (d.) 1984. Intrasite spatial analysis in Archaelogy. Cambridge University
Press, London.
HODDER, I. & C. ORTON 1976. Spatial analyis in archaeology. Cambridge University
Press, Cambridge.
JOHNSON, T. 1977. "Local Density Analysis", in Computer Applications in Archaeology
1977, pp. 90-98. University of Birmingham, Birmingham.
JOHNSON, T. 1984. "Cell frequency recording and analysis of artefact distributions".
in Hietala, H., (ed.), Intrasite spatial analysis in Archaelogy, pp. 75-96. Cambridge
University Press.
KiNTiGH, K. W. & A. J. AMMERMANN 1982. "Heuristic approaches to spatial analysis
in Archaeology", American Antiquity, 47: 31-63.
LEBART, L. 1978. "Programme d'agrgation avec constraintes (CAH contigut)". Les
Cahiers de l'analyse des donnes, III (3): 275-287.
LEROI-GOURHAN, A. & M. BREZILLON 1972. "Fouilles de PinceventEssai d'analyse
ethnographique d'un habitat magdalnien : la section 36", GALLIA Prhistoire, VII.
ORTON, C. 1982a. "Stochastic process and archaeological mechanism in spatial
analysis". Journal of Archaeological Science, 9: 1-23.
ORTON, C. R. 1982b. "Computer simulation experiments to assess the performance
of measures of quantity of pottery". World Archaeology, 14 (1): 1-20.
SiMECK, J. F. & R. R. LARICK 1981. "The recognition of multiple spatial pattern: a
case from the French upper paleolithic". Ms on fileDepartment of Anthropology.
WHALLON, R. 1973. "Spatial analysis of occupation floors I: Application of dimen-
sional analysis of variance", American Antiquity, 38: 266-278.
WHALLON, R. 1974. "Spatial analysis of occupation floors II: Application of nearest
neighbour analysis", American Antiquiiy, 39: 16-34.
WHALLON, R. 1984. "Unconstrained clustering for the analysis of spatial distributions
in archaeology", in Hietala, H., (ed.), Intrasite spatial analysis in archaeology, pp. 242-
277. Cambridge University Press.